Metabolomics Profiling of Visceral Adipose Tissue: Results From MESA and the NEO Study

Abstract

Background

Identifying associations between serum metabolites and visceral adipose tissue (VAT) could provide novel biomarkers of VAT and insights into the pathogenesis of obesity‐related diseases. We aimed to discover and replicate metabolites reflecting pathways related to VAT.

Methods and Results

Associations between fasting serum metabolites and VAT area (by computed tomography or magnetic resonance imaging) were assessed with cross‐sectional linear regression of individual‐level data from participants in MESA (Multi‐Ethnic Study of Atherosclerosis; discovery, N=1103) and the NEO (Netherlands Epidemiology of Obesity) study (replication, N=2537). Untargeted ¹H nuclear magnetic resonance metabolomics profiling of serum was performed in MESA, and metabolites were replicated in the NEO study using targeted ¹H nuclear magnetic resonance spectroscopy. A total of 30 590 metabolomic spectral variables were evaluated. After adjustment for age, sex, race/ethnicity, socioeconomic status, smoking, physical activity, glucose/lipid‐lowering medication, and body mass index, 2104 variables representing 24 nonlipid and 49 lipid/lipoprotein subclass metabolites remained significantly associated with VAT (P=4.88×10⁻²⁰–1.16×10⁻³). These included conventional metabolites, amino acids, acetylglycoproteins, intermediates of glucose and hepatic metabolism, organic acids, and subclasses of apolipoproteins, cholesterol, phospholipids, and triglycerides. Metabolites mapped to 31 biochemical pathways, including amino acid substrate use/metabolism and glycolysis/gluconeogenesis. In the replication cohort, acetylglycoproteins, branched‐chain amino acids, lactate, glutamine (inversely), and atherogenic lipids remained associated with VAT (P=1.90×10⁻³⁵–8.46×10⁻⁷), with most associations remaining after additional adjustment for surrogates of VAT (glucose level, waist circumference, and serum triglycerides), reflecting novel independent associations.

Conclusions

We identified and replicated a metabolite panel associated with VAT in 2 community‐based cohorts. These findings persisted after adjustment for body mass index and appear to define a metabolic signature of visceral adiposity.

Clinical Perspective

What Is New?

• We identified and validated a metabolite signature associated with visceral adipose tissue from a single fasting blood sample in 2 large epidemiological cohort studies.

What Are the Clinical Implications?

• These findings provide insight into potential mechanisms underpinning visceral adipose tissue metabolism distinct from generalized obesity defined by the body mass index.

• Blood‐based metabolic profiling of visceral adipose tissue using a limited set of important metabolites may address the implementation gap between recognizing the role of visceral adiposity in cardiometabolic disease and actually assessing it clinically.

Introduction

Although obesity is associated with increased risk of diabetes mellitus and cardiovascular disease (CVD), many obese individuals remain free of cardiometabolic disease.1 One factor that contributes to the heterogeneity of risk among obese individuals is the amount of visceral adipose tissue (VAT).2 Excess VAT is associated with insulin resistance, atherogenic dyslipidemia, and hepatic steatosis3; and in the long‐term, excess VAT has been linked with increases in the risk of developing type 2 diabetes mellitus4 and the metabolic syndrome,5 across the spectrum of body mass index (BMI).

Currently, precise measures of VAT are only obtainable through assessment with advanced imaging techniques, such as computed tomography and magnetic resonance imaging (MRI). Determination of VAT burden and its application to prevention or treatment of cardiometabolic outcomes are, therefore, not currently practical for routine clinical use. Anthropometric approximations, like waist circumference, are not sufficient to assess risk associated with VAT,6 and specific blood‐based metabolic markers reflecting pathways related to VAT are lacking.

The development of high‐throughput metabolomics profiling has made it feasible to acquire profiles of a whole organism's metabolic status.7 The metabolome profile can provide a high‐resolution and reproducible phenotypic signature of complex disease states, such as type 2 diabetes mellitus,8 and may offer useful biologic information that can help with understanding molecular pathways in disease. At present, there are limited data on the relationship between metabolite profiles and variation in body fat distribution, especially with VAT. Studies to date have been composed of relatively small sample sizes,9, 10 a finite number of targeted metabolites,11 or histological samples of adipose tissue alone.12 Targeting blood‐based metabolites from sufficiently large numbers of people may yield more robust and reproducible results from samples that are more easily obtained in clinical practice. Therefore, we aimed to use data from 2 large independent cohorts, MESA (Multi‐Ethnic Study of Atherosclerosis) and the NEO (Netherlands Epidemiology of Obesity) study, to discover and replicate metabolites reflecting various pathways related to visceral fat; and we performed a mendelian randomization analysis to explore the potential causal effects of atherogenic dyslipidemia on VAT deposition.

Methods

The data that support the findings of this study are available from the corresponding author on reasonable request.

Study Population and Variable Definitions

Multi‐Ethnic Study of Atherosclerosis

The overall design of MESA has been described previously.13 Briefly, MESA consists of 6814 men and women, aged 45 to 84 years, who were free of clinical CVD, of different ethnicities (white, black, Chinese American, and Hispanic) and enrolled from 6 different sites in the United States. Clinical CVD was defined as history of myocardial infarction, angina pectoris, prior revascularization, heart failure, atrial fibrillation, stroke, or peripheral arterial disease at the time of enrollment. Baseline medical history, anthropometric measurements, and laboratory data for the present study were taken from the first examination of MESA cohort (July 2000 to August 2002), as previously described.14 Education level and yearly income were determined from a self‐reported questionnaire. Physical activity was derived using a self‐reported frequency and type of leisure time physical activity and a standard conversion for metabolic equivalence units.15 Fasting serum samples from 3955 participants randomly selected were collected at the baseline visit to generate untargeted metabolomics profiles in a subset of MESA participants as part of the Development of Combinatorial Biomarkers for Subclinical Atherosclerosis initiative, a collaboration between MESA investigators and scientists at Imperial College London, as described below. At examinations 2 or 3, a random subset of 1970 MESA participants underwent abdominal computed tomography scans for aortic calcium that were subsequently used for quantifying visceral fat area: visit 2, n=756; visit 3, n=1172. For the purposes of the current study, we included 1103 participants with completed assessments of metabolomics and visceral fat. The median (interquartile range) time between metabolomics and VAT assessments was 3.2 (3.0–3.4) years.

The NEO study

The NEO study is a population‐based, prospective cohort study, including 6671 individuals aged 45 to 65 years, with an oversampling of individuals with overweight or obesity.16 Between September 2008 and 2012, men and women living in the greater area of Leiden (in the West of the Netherlands) were invited to participate if they were aged between 45 and 65 years and had a self‐reported BMI of ≥27 kg/m². In addition, all inhabitants, aged between 45 and 65 years, from one municipality (Leiderdorp) were invited to participate, irrespective of their BMI, allowing for a reference distribution of BMI. To correctly represent associations in the Dutch general population, adjustments for this oversampling have been made in the analyses17 by weighting individuals toward the BMI distribution of participants from the Leiderdorp municipality, whose BMI distribution was similar to the BMI distribution of the general Dutch population. Consequently, the results of the analyses in the NEO study apply to a population‐based study without oversampling of individuals with a BMI ≥27 kg/m².

Participants were invited to a baseline visit at the NEO study center after an overnight fast. Before this study visit, participants completed a general questionnaire at home to report demographic, lifestyle, and clinical information. At the baseline visit, an extensive physical examination was performed, including blood sampling. A high‐throughput proton nuclear magnetic resonance (NMR) metabolomics platform (Nightingale Health Ltd, Helsinki, Finland) was used to quantify 224 lipid and metabolite measures in all participants, as described below. VAT area was quantified using MRI in 2580 participants who were randomly selected from those without contraindications for MRI. After exclusion of missing data (failed MRI, n=11; failed blood sampling, n=33), 2536 participants were analyzed. Of these participants, analysis or annotation of separate metabolites failed in a median of 0.26% (interquartile range, 0.03%–3.49%), which were not imputed in the statistical analyses. Protocols were approved by the Institutional Review Board at each participating institution for MESA and by the Medical Ethical Committee of the Leiden University Medical Center for the NEO study. All participants provided written informed consent.

Metabolomics Measurements

In MESA (the discovery cohort), untargeted ¹H NMR analysis of serum samples obtained at the baseline examination was performed using a method previously described.18 MESA samples used in the current study were analyzed in 2 phases as part of the European Union–funded Development of Combinatorial Biomarkers for Subclinical Atherosclerosis project. Details about the preparation of samples, including quality controls, NMR data acquisition, and NMR data processing, are described in Data S1. The specific NMR data sets used in the current study include the following: (1) standard 1‐dimensional NMR spectrum showing resonances from all proton‐containing molecules in the sample, including broad, largely undefined bands from serum proteins, sharper and well‐defined bands from serum lipoproteins (with some classification into their main groups), and sharp peaks from a range of small‐molecule metabolites, such as amino acids, simple carbohydrates, organic acids, organic bases, and several osmolytes; (2) Carr‐Purcell‐Meiboom‐Gill spectrum that attenuates the peaks from the macromolecules and allows better definition of the small molecules; and (3) quantification of lipoprotein subclasses obtained from deconvolution of the methyl peak near δ0.89 using a Bruker (Bruker Biospin, Rheinstetten, Germany) procedure adapted from the method of Petersen et al.19 Bruker NMR measurements included total high‐density lipoprotein (HDL), low‐density lipoprotein (LDL), triglycerides, and cholesterol as well as analysis of 105 lipoprotein subclasses, including different chemical components of intermediate‐density lipoprotein (density, 1.006–1.019 kg/L), very‐LDL (VLDL; density, 0.950–1.006 kg/L), LDL (density, 1.09–1.63 kg/L), and HDL (density, 1.063–1.210 kg/L). The LDL subfraction was separated into 6 density classes (LDL‐1, 1.019–1.031 kg/L; LDL‐2, 1.031–1.034 kg/L; LDL‐3, 1.034–1.037 kg/L; LDL‐4, 1.037–1.040 kg/L; LDL‐5, 1.040–1.044 kg/L; and LDL‐6, 1.044–1.063 kg/L), and the HDL subfraction was separated into 4 density classes (HDL‐1, 1.063–1.100 kg/L; HDL‐2, 1.100–1.125 kg/L; HDL‐3, 1.125–1.175 kg/L; and HDL‐4, 1.175–1.210 kg/L). These specific NMR spectra have been previously tested for quality control, harmonization, and alignment.20

In the NEO study (the replication cohort), targeted metabolomics were measured using a high‐throughput proton NMR metabolomics platform (Nightingale Health Ltd) to quantify 224 lipid and metabolite measures in all participants. The NMR spectroscopy was conducted at the Medical Research Council Integrative Epidemiology Unit at the University of Bristol (Bristol, UK) and processed by Nightingale's biomarker quantification algorithms (version 2014). This method provides quantification of lipoprotein subclass profiling with lipid concentrations within 14 lipoprotein subclasses. The 14 subclass sizes were defined as follows: extremely large VLDL with particle diameters from 75 nm upwards and a possible contribution of chylomicrons, 5 VLDL subclasses (average particle diameters, 64.0, 53.6, 44.5, 36.8, and 31.3 nm), intermediate‐density lipoprotein (average particle diameter, 28.6 nm), 3 LDL subclasses (average particle diameters, 25.5, 23.0, and 18.7 nm), and 4 HDL subclasses (average particle diameters, 14.3, 12.1, 10.9, and 8.7 nm). Within the lipoprotein subclasses, the following components were quantified: total cholesterol, total lipids, phospholipids, free cholesterol, cholesteryl esters, and triglycerides. The mean size for VLDL, LDL, and HDL particles was calculated by weighting the corresponding subclass diameters with their particle concentrations. Furthermore, 58 metabolic measures were determined that belong to classes of apolipoproteins, cholesterol, fatty acids, glycerides, phospholipids, amino acids, fluid balance, glycolysis‐related metabolites, inflammation, and ketone bodies. Details of the experimentation and applications of the NMR metabolomics platform have been described previously,21 as well as CVs (coefficients of variation) for metabolic biomarkers.22, 23 A full list of the measured biomarkers in the NEO study is included in Table S1.

Body Fat and VAT Measurements

In MESA, weight and height were measured using a balance‐beam scale and stadiometer, respectively, and used to calculate BMI as weight (in kilograms) divided by height (in meters) squared. Waist circumference was measured at the minimum abdominal girth using a steel measuring tape of standard 4‐ounce tension in centimeters. Electron‐beam or multidetector computed tomography scans of the abdomen, obtained to measure aortic calcification, were used to measure fat and lean area in the abdomen, as previously described.5 Briefly, visceral fat was defined as the fat enclosed by the visceral cavity, and fat tissue was identified as being between −190 and −30 Hounsfield units. Within the area of interest, the density value was assigned to each pixel using the MIPAV 4.1.2 software (National Institutes of Health, Bethesda, MD) as fat or lean tissue. Six transverse cross‐sectional slices were analyzed (2 at L2–3, 2 at L3–4, and 2 at L4–5), and visceral fat area (cm²) was calculated as the average of the sum of visceral fat over all 6 available slices. Interreader and intrareader reliability for visceral fat area was 0.99 for all measures.

In the NEO study, body weight was assessed by the Tanita bioimpedance balance (TBF‐310; Tanita International Division, UK) without shoes, and 1 kg was subtracted from the body weight. Waist circumference was measured midway between the border of the lower costal margin and the iliac crest. Abdominal visceral fat was quantified by a turbo spin echo imaging protocol using MRI. Imaging was performed on a 1.5‐T MR system (Philips Medical Systems, Best, the Netherlands). At the level of the fifth lumbar vertebra, 3 transverse images, each with a slice thickness of 10 mm, were obtained during a breath hold. Visceral fat area was quantified by converting the number of pixels to square centimeters for all 3 slides, and the mean area of the 3 slides was used in the analyses. Earlier studies have shown that such cross‐sectional images are highly correlated to total volumes (correlation coefficients, ≈0.8) and can, therefore, validly represent VAT.24

Statistical Analysis

Baseline characteristics of the study populations are presented as median (interquartile range) or proportion (percentage), as appropriate. Multivariable linear regression models were constructed to assess the association of metabolites with VAT for all NMR experiments. To allow for comparisons, metabolites were logarithmically transformed and standardized to a mean of 0 and an SD of 1. VAT was confirmed to be normally distributed. Linear regression modeling was performed, with the metabolite as the exposure variable and mean VAT area as the outcome variable, on the basis of a hypothesis‐free design because the biological features of metabolomics and VAT may be bidirectional (ie, metabolites may influence VAT accumulation/function, and/or VAT accumulation may influence downstream metabolic processes). Furthermore, to uncover potential causal pathways to visceral fat accumulation, we also performed mendelian randomization analyses with the replicated metabolites, where possible (see method below). Models were adjusted for age, sex, race/ethnicity, socioeconomic status, smoking, physical activity, glucose and lipid‐lowering medication use, and BMI to investigate to what extent the associations were specific for VAT and not merely overall body mass. Given the hypothesis‐free design and the large number of comparisons, we adjusted for multiple testing using a predefined false‐discovery rate threshold of 1% for the primary analysis. Given known differences in body fat distribution by sex and race/ethnicity, secondary analyses were performed, stratified by these variables. We also performed targeted pathway analysis using Metaboanalyst (http://www.metaboanalyst.ca), a web‐based tool for metabolomics analysis, and interpretation that uses the Kyoto Encyclopedia of Genes and Genomes and Small Molecule Pathway databases to perform overrepresentation, pathway enrichment, and pathway topological analyses (explained in Data S1). They were used to determine the overall associations of our metabolite set that map to particular pathways related to VAT and assess whether the metabolites are critical connectors within the pathways’ network structure.25

Next, we used the NEO study as a separate cohort to replicate our findings with the same statistical analysis strategy on a targeted metabolomics platform. All analyses in the NEO study were weighted toward the BMI distribution of the general population. A predefined false‐discovery rate threshold of 1% was also used for this analysis. Using the replicated metabolites, to identify novel VAT‐associated metabolites beyond known correlates, we additionally sequentially adjusted for the following: (1) fasting glucose concentrations and waist circumference; and (2) plasma triglyceride concentrations, to investigate if and what metabolites remained after adjustment for additional modifiers of metabolic disease and indirect surrogate markers for VAT (eg, “hypertriglyceridemic waist”).26 Finally, to better understand the potential directionality of the association between lipid‐based metabolites and VAT (ie, does dyslipidemia influence VAT deposition), we estimated the potential causal effects of overall measures of HDL cholesterol (HDL‐C), LDL cholesterol, and triglycerides on VAT volume by performing 2‐sample mendelian randomization analyses using genetic instruments linked to blood lipid levels and combining the summary statistics of large genome‐wide meta‐analyses on blood lipid levels and VAT (explained in Data S1). Statistical analyses were performed using SAS software, version 9.4 (SAS Corporation, Cary, NC), and Stata Statistical Software, version 14.0 (Statacorp, College Station, TX).

Results

Characteristics of the discovery and replication study cohorts are presented in Table 1. Both cohorts were primarily middle aged, with ≈50% women. MESA cohort was racially/ethnically diverse, with ≈60% nonwhite participants, compared with the NEO study cohort, which was predominantly white. The median BMIs, waist circumferences, and VAT areas for women and men were modestly higher in MESA than in the NEO study, generally reflecting known demographic and anthropometric differences between the United States and the Netherlands.

Table 1.

Baseline Characteristics of the Study Populations

Clinical Characteristics	MESA (n=1103)	NEO Study (n=2536)
Demographics
Age, y	63.0 (54.0–70.0)	56.0 (51.0–61.0)
Men, %	51.6	47.5
Race/ethnicity, %
White	39.8	95.9
Black	17.6	…
Hispanic	27.9	…
Chinese	14.7	…
Other	…	4.1
Education level, %
Low (some or graduated high school)	35.5	53.7
High (vocational school, university, and postgraduate)	64.5	46.3
Income, $/y, %
0–34 999	43.9	N/A
35 000–99 999	39.8	N/A
≥100 000	16.4	N/A
Medical history
Hypertension, %	48.5	19.7
Diabetes mellitus, %	11.4	3.3
Dyslipidemia, %	43.9	42.5
Metabolic syndrome, %	34.5	23.7
Current smoker, %	14.1	14.4
Moderate and vigorous physical activity, MET×min/wk	4001.3 (2032.5–7260.0)	2850.0 (1597.5–4905.0)
Systolic BP, mm Hg	124.0 (111.0–141.0)	129.0 (118.0–141.0)
Diastolic BP, mm Hg	72.0 (65.0–79.0)	83.0 (76.0–90.0)
BP ≥130/85 mm Hg, %	36.0	56.0
Triglycerides, mg/dL	119.0 (79.0–175.0)	90.3 (64.6–131.9)
Triglycerides ≥150 mg/dL, %	35.1	19.0
HDL‐C, mg/dL	48.0 (40.0–59.0)	57.9 (47.5–71.4)
HDL‐C <40 mg/dL (men) or <50 mg/dL (women), %	35.8	15.4
Fasting glucose, mg/dL	91.0 (84.0–99.0)	95.3 (89.7–102.5)
Fasting glucose ≥100 mg/dL, %	24.9	32.3
Body composition
BMI, kg/m2	Women: 27.3 (24.4–31.3)	Women: 24.9 (22.0–27.5)
	Men: 27.2 (24.4–30.1)	Men: 26.3 (24.2–28.5)
Waist circumference, cm	Women: 96.0 (85.8–105.1)	Women: 84.0 (77.0–94.0)
	Men: 97.5 (90.6–106.3)	Men: 97.0 (91.0–104.0)
Waist circumference ≥102 cm (men) or ≥88 cm (women), %	Women: 70.0	Women: 38.0
	Men: 36.4	Men: 32.6
VAT area, cm2	Women: 122.4 (82.1–183.0)	Women: 56.8 (36.6–89.0)
	Men: 191.6 (128.0–248.3)	Men: 105.6 (75.1–144.2)

Data are presented as median (interquartile range) or proportion (percentage), as appropriate. Results from the NEO study are based on analyses weighted toward the BMI distribution of the general population. Number of missing values per variable in the NEO study: ethnicity, 4; education, 26; hypertension, 6; diabetes mellitus, 7; metabolic syndrome, 7; smoking, 3; physical activity, 11; diastolic BP, 1; triglycerides, 6; HDL‐C, 6; and fasting glucose, 9 (no missing values for other variables). BMI indicates body mass index; BP, blood pressure; HDL‐C, high‐density lipoprotein cholesterol; MESA, Multi‐Ethnic Study of Atherosclerosis; MET, metabolic equivalent; N/A, not applicable; NEO, Netherlands Epidemiology in Obesity; VAT, visceral adipose tissue.

Metabolite Profiling in MESA

In MESA discovery cohort, 30 590 metabolomic spectral variables were evaluated in untargeted metabolomics analyses using NMR. After multivariable adjustment for age, sex, race/ethnicity, socioeconomic status, smoking, physical activity, glucose and lipid‐lowering medication use, and BMI, 2104 spectral variables representing 24 nonlipid (Table 2) and 49 lipid/lipoprotein subclass metabolites (Table 3) remained statistically significantly associated with VAT (P=4.88×10⁻²⁰–1.16×10⁻³). These included conventional clinical metabolites (eg, creatinine), amino acids and their by‐products (eg, leucine, isoleucine, glutamine [inversely associated], valine, and proline), acetylglycoproteins and mannose, intermediates of glucose and hepatic metabolism (eg, glycerol, glucose, and choline), organic acids (eg, lactate), subclasses of very‐low‐density, low‐density, intermediate‐density, and high‐density apolipoproteins, cholesterol, phospholipids, and triglycerides. In general, among the lipid‐based metabolites, HDL‐related metabolites were inversely associated with VAT. Conversely, intermediate‐density lipoprotein and VLDL particles were almost uniformly positively associated with VAT. Metabolite profiles were generally consistent between men and women and between white and nonwhite participants in stratified analyses (Figures S1 and S2).

Table 2.

Associations Between Nonlipid Metabolites and VAT

MESA			NEO Study
Metabolite	Effect Estimate β (95% CI)	Nominal P Value	Metabolite	Effect Estimate β (95% CI)	Nominal P Value
1‐Dimensional NMR
Acetylglycoproteins	14.50 (10.87 to 18.13)	1.14E‐14	Acetylglycoproteins	11.70 (9.86 to 13.54)	1.58E‐34a
Choline	−15.54 (−19.42 to −11.66)	9.52E‐15	···	···	···
Creatinine	12.88 (9.30 to 16.45)	2.97E‐12	Creatinine	1.21 (−0.75 to 3.16)	2.25E‐01
Glycerol	8.43 (4.65 to 12.20)	1.33E‐05	···	···	···
Glyceryl groups of lipids	13.83 (10.29 to 17.36)	4.15E‐14	···	···	···
Lactate	13.73 (10.14 to 17.32)	1.38E‐13	Lactate	4.75 (2.86 to 6.63)	8.46E‐07a
Mannose	15.92 (12.33 to 19.52)	1.49E‐17	···	···	···
Myoinositol	7.96 (4.24 to 11.69)	3.05E‐05	···	···	···
Proline	12.90 (9.21 to 16.59)	1.26E‐11	···	···	···
Carr‐Purcell‐Meiboom‐Gill echo acquisition
2‐Ketoisovalerate	3.35 (2.24 to 4.46)	4.73E‐09	···	···	···
Acetylglycoproteins	9.22 (6.81 to 11.63)	1.41E‐13	Acetylglycoproteins	11.70 (9.86 to 13.54)	1.58E‐34a
Alanine	−3.50 (−4.68 to −2.33)	5.98E‐09	···	···	···
Albumin	−1.88 (−2.40 to −1.35)	5.98E‐12	Albumin	−0.02 (−1.80 to 1.76)	9.84E‐01
α‐Glucose	−8.34 (−11.40 to −5.27)	1.19E‐07	···	···	···
Arginine	1.01 (0.59 to 1.43)	3.25E‐06	···	···	···
β‐Glucose	−3.29 (−4.47 to −2.11)	5.99E‐08	···	···	···
Choline	−6.45 (−8.56 to −4.34)	3.00E‐09	···	···	···
Citrate	−0.30 (−0.47 to −0.14)	2.79E‐04	···	···	···
Creatinine	2.99 (2.19 to 3.78)	3.74E‐13	Creatinine	1.21 (−0.75 to 3.16)	2.25E‐01
Ornithine	−1.62 (−2.41 to −0.84)	5.64E‐05	···	···	···
Glutamate	0.33 (0.14 to 0.51)	5.04E‐04	···	···	···
Glutamine	−1.63 (−2.24 to −1.03)	1.42E‐07	Glutamine	−3.09 (−5.05 to −1.13)	2.01E‐03a
Glyceryl groups of lipids	2.02 (1.54 to 2.50)	4.18E‐16	···	···	···
Isoleucine	2.50 (1.88 to 3.12)	5.25E‐15	Isoleucine	13.22 (11.16 to 15.28)	3.78E‐35a
Lactate	6.44 (4.81 to 8.07)	2.44E‐14	Lactate	4.75 (2.86 to 6.63)	8.46E‐07a
Leucine	3.65 (2.50 to 4.79)	7.03E‐10	Leucine	12.58 (10.36 to 14.80)	5.23E‐28a
Lysine	−8.96 (−10.94 to −6.98)	3.25E‐18	···	···	···
Mannose	11.42 (9.02 to 13.81)	4.88E‐20	···	···	···
Proline	5.54 (4.12 to 6.96)	5.47E‐14	···	···	···
Pyroglutamate	−1.05 (−1.36 to −0.73)	1.15E‐10	···	···	···
Valine	3.05 (1.95 to 4.15)	6.55E‐08	Valine	6.89 (4.68 to 9.10)	1.07E‐09a

Model adjusted for age, sex, race/ethnicity, socioeconomic status, smoking, physical activity, glucose and lipid‐lowering medication use, and body mass index. Effect estimate β represents the difference in VAT area (in cm²) per 1‐SD in metabolite intensity (relative units). MESA indicates Multi‐Ethnic Study of Atherosclerosis; NEO, Netherlands Epidemiology in Obesity; NMR, nuclear magnetic resonance; VAT, visceral adipose tissue.

^aMetabolites that were significant in the NEO study data set after false‐discovery rate correction.

Table 3.

Associations Between Lipid Metabolites and VAT

MESA			NEO Study
Metabolite	Effect Estimate β (95% CI)	Nominal P Value	Metabolite	Effect Estimate β (95% CI)	Nominal P Value
HDL cholesterol	−11.00 (−14.74 to −7.25)	1.14E‐08	HDL cholesterol	−8.11 (−10.21 to −6.00)	5.67E‐14a
HDL free cholesterol	−13.21 (−17.15 to −9.27)	7.81E‐11	···	···	···
HDL phospholipids	−11.50 (−15.50 to −7.50)	2.22E‐08	···	···	···
Total plasma apolipoprotein‐A1	−7.59 (−11.36 to −3.81)	8.84E‐05	Total plasma apolipoprotein‐A1	−2.94 (−5.04 to −0.84)	6.09E‐03a
Extralarge HDL apolipoprotein‐A1	−10.92 (−14.75 to −7.10)	2.82E‐08	···	···	···
Extralarge HDL cholesterol	−10.72 (−14.56 to −6.89)	5.47E‐08	Extralarge HDL cholesterol	−7.52 (−9.61 to −5.44)	1.99E‐12a
Extralarge HDL free cholesterol	−12.88 (−16.61 to −9.15)	2.18E‐11	Extralarge HDL free cholesterol	−8.24 (−10.34 to −6.13)	2.49E‐14a
Extralarge HDL phospholipids	−12.78 (−16.92 to −8.63)	2.12E‐09	Extralarge HDL phospholipids	−10.66 (−12.83 to −8.49)	1.65E‐21a
Large HDL apolipoprotein‐A1	−8.90 (−13.38 to −4.41)	1.07E‐04	···	···	···
Large HDL cholesterol	−10.68 (−14.48 to −6.89)	4.33E‐08	Large HDL cholesterol	−10.99 (−13.07 to −8.92)	8.21E‐25a
Large HDL free cholesterol	−12.72 (−16.66 to −8.78)	3.76E‐10	Large HDL free cholesterol	−10.98 (−13.00 to −8.95)	9.22E‐26a
Large HDL phospholipids	−10.90 (−14.80 to −6.99)	5.66E‐08	Large HDL phospholipids	−9.52 (−11.67 to −7.37)	7.09E‐18a
Medium HDL cholesterol	−9.15 (−12.95 to −5.35)	2.70E‐06	Medium HDL cholesterol	−3.58 (−5.63 to −1.54)	5.92E‐04a
Medium HDL free cholesterol	−9.83 (−14.01 to −5.65)	4.56E‐06	Medium HDL free cholesterol	−3.52 (−5.61 to −1.44)	9.37E‐04a
Medium HDL phospholipids	−7.88 (−11.78 to −3.97)	8.33E‐05	Medium HDL phospholipids	−1.61 (−3.68 to 0.46)	1.27E‐01
Medium HDL triglycerides	6.44 (2.56 to 10.31)	1.16E‐03	Medium HDL triglycerides	8.65 (6.37 to 10.92)	1.26E‐13a
Small HDL triglycerides	11.31 (7.78 to 14.84)	4.87E‐10	Small HDL triglycerides	10.65 (8.91 to 12.39)	2.31E‐32a
IDL apolipoprotein‐B	7.22 (3.36 to 11.08)	2.63E‐04	···	···	···
IDL cholesterol	7.03 (3.27 to 10.80)	2.68E‐04	IDL cholesterol	2.36 (0.26 to 4.46)	2.79E‐02
IDL free cholesterol	6.96 (3.17 to 10.76)	3.41E‐04	IDL free cholesterol	0.10 (−1.97 to 2.18)	9.22E‐01
IDL phospholipids	9.29 (5.47 to 13.11)	2.18E‐06	IDL phospholipids	2.10 (0.04 to 4.16)	4.56E‐02
IDL triglycerides	11.42 (7.59 to 15.25)	6.89E‐09	IDL triglycerides	7.10 (5.50 to 8.70)	6.03E‐18a
LDL triglycerides	6.68 (3.08 to 10.28)	2.89E‐04	LDL triglycerides	5.59 (3.92 to 7.26)	6.52E‐11a
LDL‐3 free cholesterol	−8.34 (−12.35 to −4.34)	4.83E‐05	···	···	···
LDL‐5 triglycerides	6.56 (2.91 to 10.21)	4.46E‐04	···	···	···
Total triglycerides	14.28 (10.60 to 17.96)	6.07E‐14	Total triglycerides	11.10 (9.38 to 12.83)	1.90E‐35a
VLDL apolipoprotein‐B	12.48 (8.81 to 16.15)	4.37E‐11	···	···	···
VLDL cholesterol	10.86 (7.21 to 14.52)	7.85E‐09	VLDL cholesterol	8.77 (6.84 to 10.71)	1.22E‐18a
VLDL free cholesterol	12.87 (9.20 to 16.54)	1.07E‐11	···	···	···
VLDL phospholipids	13.43 (9.75 to 17.12)	1.76E‐12	···	···	···
VLDL triglycerides	14.91 (11.21 to 18.61)	7.55E‐15	VLDL triglycerides	11.39 (9.63 to 13.15)	8.09E‐36a
XXL VLDL cholesterol	10.29 (6.41 to 14.16)	2.40E‐07	XXL VLDL cholesterol	7.18 (4.65 to 9.71)	3.05E‐08a
XXL VLDL free cholesterol	11.68 (7.97 to 15.39)	9.89E‐10	XXL VLDL free cholesterol	8.09 (5.60 to 10.58)	2.34E‐10a
XXL VLDL phospholipids	15.30 (11.63 to 18.96)	8.45E‐16	XXL VLDL phospholipids	9.13 (6.10 to 12.15)	3.81E‐09a
XXL VLDL triglycerides	16.17 (12.52 to 19.83)	1.78E‐17	XXL VLDL triglycerides	9.37 (4.55 to 14.20)	1.43E‐04a
Extralarge VLDL cholesterol	10.23 (6.59 to 13.87)	4.53E‐08	Extralarge VLDL cholesterol	7.33 (4.57 to 10.09)	2.03E‐07a
Extralarge VLDL free cholesterol	9.82 (6.11 to 13.52)	2.52E‐07	Extralarge VLDL free cholesterol	7.31 (4.75 to 9.87)	2.28E‐08a
Extralarge VLDL phospholipids	13.22 (9.56 to 16.88)	2.80E‐12	Extralarge VLDL phospholipids	7.67 (4.39 to 10.94)	4.63E‐06a
Extralarge VLDL triglycerides	12.92 (9.25 to 16.58)	8.77E‐12	Extralarge VLDL triglycerides	9.04 (5.14 to 12.94)	5.87E‐06a
Large VLDL cholesterol	10.34 (6.68 to 14.00)	4.00E‐08	Large VLDL cholesterol	9.82 (7.84 to 11.79)	4.35E‐22a
Large VLDL free cholesterol	10.58 (6.89 to 14.28)	2.51E‐08	Large VLDL free cholesterol	9.82 (7.90 to 11.73)	2.50E‐23a
Large VLDL phospholipids	11.85 (8.18 to 15.52)	3.75E‐10	Large VLDL phospholipids	10.54 (8.55 to 12.53)	9.34E‐25a
Large VLDL triglycerides	11.32 (7.65 to 14.98)	2.12E‐09	Large VLDL triglycerides	11.24 (9.23 to 13.24)	2.21E‐27a
Medium VLDL cholesterol	8.18 (4.60 to 11.75)	8.24E‐06	Medium VLDL cholesterol	9.88 (7.95 to 11.81)	2.52E‐23a
Medium VLDL free cholesterol	8.04 (4.42 to 11.67)	1.52E‐05	Medium VLDL free cholesterol	10.99 (9.16 to 12.82)	3.37E‐31a
Medium VLDL phospholipids	9.51 (5.89 to 13.13)	3.16E‐07	Medium VLDL phospholipids	11.09 (9.28 to 12.90)	2.52E‐32a
Medium VLDL triglycerides	10.08 (6.43 to 13.74)	8.02E‐08	Medium VLDL triglycerides	11.39 (9.57 to 13.20)	1.07E‐33a
Extrasmall VLDL cholesterol	−8.38 (−11.94 to −4.82)	4.58E‐06	···	···	···
Extrasmall VLDL phospholipids	12.18 (8.55 to 15.82)	8.37E‐11	Extrasmall VLDL phospholipids	4.62 (2.67 to 6.57)	3.53E‐06a

Model adjusted for age, sex, race/ethnicity, socioeconomic status, smoking, physical activity, glucose and lipid‐lowering medication use, and body mass index. Effect estimate β represents the difference in VAT area (in cm²) per 1‐SD in metabolite intensity (relative units). Lipoprotein particle subclasses range in size from extrasmall to XXL. HDL indicates high‐density lipoprotein; IDL, intermediate‐density lipoprotein; LDL, low‐density lipoprotein; MESA, Multi‐Ethnic Study of Atherosclerosis; NEO, Netherlands Epidemiology in Obesity; VAT, visceral adipose tissue; VLDL, very‐LDL; XXL, very extralarge.

^aMetabolites that were significant in the NEO study data set after false‐discovery rate correction.

Pathway analyses were performed using overrepresentation, pathway enrichment, and pathway topological analysis methods for the nonlipid metabolites. Thirty‐one distinct biochemical pathways were identified, mapping to the metabolite set significantly associated with VAT (Figure 1). The pathways with the strongest associations with visceral adiposity (based on P values derived from pathway enrichment analyses reflecting the overall association of the metabolite set) included those using amino acids as substrates for biosynthetic processes, such as aminoacyl‐tRNA biosynthesis (P=3.39×10⁻¹⁰) and branched‐chain amino acid degradation (P=1.30×10⁻⁴). Other pathways included metabolism of other amino acids and glycolysis/gluconeogenesis. A full list of the metabolic pathways associated with visceral adiposity and centrality/impact of the metabolites on each specific pathway is given in Table S2.

Figure 1.

Targeted metabolomics pathway analysis in MESA (Multi‐Ethnic Study of Atherosclerosis). Each node represents a separate biochemical pathway. The color of the node corresponds to its location on the y axis and indicates statistical significance in terms of ‐log(P) (higher values correspond to lower P values; eg, red nodes have low P values and yellow nodes have high P values). P values are derived from pathway enrichment analyses that measure the overall association of a set of metabolites that map to a particular pathway with the phenotype being examined (visceral adiposity). The size of the node corresponds to its location on the x axis and indicates to some extent the centrality of the metabolites in the data set for the represented pathway. This “pathway impact” measure combines theoretic measures to suggest whether the metabolites are critical connectors within a network as opposed to being more peripheral nodes. The total pathway impact for all metabolites in any given pathway from the metabolome databases (eg, Kyoto Encyclopedia of Genes and Genomes and Small Molecule Pathway databases) sum to 1. The pathway impact reported herein is the cumulative total of pathway impact for all metabolites used for analysis.

Replication Analysis: The NEO Study

To replicate our findings from MESA in a different epidemiological cohort, we repeated the analyses with the metabolites that were significantly associated with VAT in the discovery cohort by using the targeted Nightingale metabolomics platform in the NEO study cohort. In this analysis, 6 of the nonlipid (Table 2) and 34 of the lipid/lipoprotein subclass metabolites (Table 3) were replicated and retained statistical significance in the NEO study using a prespecified false‐discovery rate 1% threshold. The β coefficients (reflecting the magnitude of association between metabolites and VAT) were highly correlated between MESA and the NEO study (R ²=0.68, Figure 2). Unadjusted correlations between adiposity variables and replicated metabolites in both MESA and the NEO study are reported in Table S3. Similar patterns for metabolite‐VAT associations in sex‐ and race/ethnicity‐stratified analyses were seen in the replication cohort as in the discovery cohort (Figures S1 and S2).

Figure 2.

Associations between metabolites and visceral adipose tissue: correlation of the β coefficients between the 2 cohort studies. Scatterplot with regression line of β coefficients from each cohort study with each colored dot representing an individual metabolite. β Coefficients represent the difference in visceral adipose tissue area (in cm²) per SD metabolite intensity and are from a model adjusted for age, sex, race/ethnicity, socioeconomic status, smoking, physical activity, glucose and lipid‐lowering medication use, and body mass index. HDL indicates high‐density lipoprotein; IDL, intermediate‐density lipoprotein; LDL, low‐density lipoprotein; MESA, Multi‐Ethnic Study of Atherosclerosis; NEO, Netherlands Epidemiology in Obesity; VLDL, very‐LDL.

Among the replicated metabolites (selecting HDL‐C, VLDL cholesterol, and serum triglycerides to represent the broad categories of related lipids/lipoproteins associated with VAT), we performed sequential adjustment for fasting glucose concentrations and waist circumference and found the associations between the selected replicated metabolites and VAT were slightly weaker but retained statistical significance (Figure 3). After further adjustment for plasma triglyceride concentrations (accounting for hypertriglyceridemic waist), acetylglycoproteins, branched‐chain amino acids (isoleucine, leucine, and valine), glutamine (inversely), and serum triglycerides remained significantly associated with VAT (nominal P<0.05 for all, Figure 3).

Figure 3.

Associations between selected metabolites and visceral adiposity, adjusted for important metabolic phenotypes in the NEO (Netherlands Epidemiology in Obesity) study. Forest plot of associations between selected metabolites and visceral adipose tissue in the NEO study cohort. Each set of 3 nodes on the graph corresponds to a different metabolite. The first (red) node in each set represents the difference with 95% CI in visceral adipose tissue (VAT) area (in cm²) per 1‐SD metabolite intensity, adjusted for age, sex, race/ethnicity, socioeconomic status, smoking, physical activity, glucose and lipid‐lowering medication use, and body mass index. The second (orange) node in each set represents the model additionally adjusted for fasting plasma glucose level and waist circumference, and the third (blue) node represents the model additionally adjusted for serum triglyceride level, measured by standard assay.

Mendelian Randomization Study

Two‐sample mendelian randomization analyses using genetic instruments for blood lipid levels were performed by combining the summary statistics of large‐scale genome‐wide meta‐analyses on blood lipid levels and VAT. Data on both the instrument‐exposure (blood lipid levels) and instrument‐outcome (VAT volume) associations were available for 208 instruments (HDL‐C, n=83; LDL cholesterol, n=72; triglycerides, n=53; with 9 serving as instruments for multiple traits), after harmonization. As shown in Figure S3, we did not find evidence for a causal effect of overall measures of HDL‐C, LDL cholesterol, and triglyceride blood levels on VAT volume using the assessed genetic instruments linked to blood lipid levels.

Discussion

Using an untargeted metabolomics platform and a comprehensive pathway analysis tool in a large, multiethnic population cohort (MESA), we identified a metabolite signature associated with VAT linked to several putative biological pathways, including amino acid substrate use/metabolism and glycolysis/gluconeogenesis. We then replicated our findings in a separate epidemiological cohort (NEO study) using targeted metabolomics and found that acetylglycoproteins, branched‐chain amino acids (isoleucine, leucine, and valine), glutamine (inversely), and serum triglycerides by ¹H NMR remained associated with VAT, even after adjustment for established surrogate biomarkers of VAT (BMI, fasting glucose, waist circumference, and serum triglycerides), suggesting that a single, fasting measurement of metabolites can provide biological information beyond standard risk markers of visceral fat. We believe these findings provide insight into potential mechanisms underpinning VAT metabolism distinct from generalized obesity (defined by BMI) and help to define a metabolic signature of visceral adiposity.

A growing number of studies have used targeted metabolic profiling as a tool for biomarker discovery in obesity, but studies to date have been composed of relatively small sample sizes9, 10 or histological samples of adipose tissue alone,12 without targeting plasma‐based metabolites that may be more easily obtained in clinical practice. Menni and colleagues11 performed targeted metabolomics profiling of 208 plasma metabolites on 2401 women in the United Kingdom and assessed their relation to VAT measured by dual x‐ray absorptiometry. They also observed associations between branched‐chain amino acids, lactate, and VAT but did not perform replication studies to confirm their findings. Thus, one of the strengths of our investigation is the use of 2 well‐characterized prospective cohorts, 1 for derivation and 1 for replication, each with dedicated imaging assessments of VAT, rather than relying on surrogate markers of VAT, such as anthropometric measurements. Furthermore, we use robust untargeted NMR‐based experiments initially to broadly characterize the metabolic phenotype related to VAT and then replicate our findings using a targeted NMR approach in a cohort that is well diversified demographically and geographically from the derivation cohort. All individuals in our study had assessments of BMI, waist circumference, and fasting glucose and triglycerides, allowing us to adjust for overall adiposity, glucose intolerance, and dyslipidemia.

Several limitations of the study merit comment. First, our findings should be primarily understood within a biological context; the utility of these metabolites for use in predictive modeling when added to standard clinical risk scores requires further study. Second, for MESA, because the metabolites were measured at a different time point than the abdominal imaging, we cannot exclude the possibility that metabolite concentrations might have differed at the follow‐up examination. Different imaging methods were used to estimate VAT in each cohort; however, the imaging for both cohorts included the area around the fifth lumbar vertebrae, and multiple transverse cross‐sectional slices were analyzed and averaged to obtain the final mean VAT value comparable between cohorts. Furthermore, prior work showed good agreement (<3% difference in Bland‐Altman analysis) between computed tomography and MRI for the measurement of VAT.27 Although the cohorts varied both geographically and demographically, and metabolites in each cohort were measured using different algorithms, the replication observed across cohorts despite these differences in study populations (different amounts of VAT, different demographics, and different metabolomics platforms) makes our findings robust. However, it is possible that differences in ethnicity, diet, or distribution of obesity between the cohorts could partially explain the variability observed in metabolite associations in race‐stratified analyses. These differences are likely most important for lipid metabolites given the known differences in lipid profiles between white and black individuals.28 Furthermore, these differences may at least partially explain the observation that some metabolites found to be significant in MESA are not replicated in the NEO study. Moreover, we cannot generalize to other populations not well represented in either cohort in which alternative metabolite relationships may exist. Because our study was cross‐sectional by design, we cannot comment on the relationship between temporal changes in metabolite levels and visceral fat. However, although the Mendelian randomization analyses did not demonstrate a causal relationship between genetic instruments linked to blood lipid levels and VAT volume, a reverse directionality is more likely in that excess VAT may cause an atherogenic dyslipidemia. In line with the study by Xu et al,29 because of the large differences in GWAS (genome‐wide association study) sample size (n=322 154 for BMI, and n=18 832 for visceral fat), we cannot exclude that small causal effects of blood lipid concentrations on visceral fat may have been undetected. Further Mendelian randomization studies using genetic instruments linked to VAT will help elucidate the causal effects of VAT on metabolic traits. Finally, although we identified several biological pathways using metabolites associated with VAT, our interpretation of pathways must remain circumspect and hypothesis generating. In many instances, the identified metabolites represented substrates in the pathway rather than products, yielding one‐sided evidence of biological relevance. Furthermore, the level of metabolomics detail derived with 1H NMR is not sufficient to yield firm conclusions about the involvement of pathways.

Our findings, which highlight acetylglycoproteins, branched‐chain amino acids, lactate, glutamine (inversely associated), and an atherogenic dyslipidemic profile (high triglycerides and VLDL and low HDL) from hundreds of metabolites assayed, are noteworthy in the context of experimental and clinical data suggesting that certain metabolites may be both markers and mediators of adverse health outcomes related to visceral obesity. For example, breakdown products of acetylglycoproteins, such as mannose, are elevated in individuals with insulin resistance30 and associated with incident type 2 diabetes mellitus and CVD.31 Indeed, we found that total acetylglycoproteins (and mannose in MESA) were significantly positively associated with VAT and that they remained associated with VAT even after adjustment for markers of glycemia and dyslipidemia in the NEO study. Acetylglycoproteins may perform a variety of cellular functions, including enzymatic catalysis, protein folding, conformation, and stabilization of biological membranes important for metabolic homeostasis32; perturbation of this highly regulated system may increase circulating concentrations of acetylglycoproteins and represent a potential biomarker of visceral adiposity‐related disease risk.

Glutamine, the most abundant free amino acid in human blood,33 plays a role in a variety of biochemical functions and was inversely associated with VAT in our study. In prior work, urinary glutamine was inversely related to higher BMI and waist circumference in a population‐based sample of adults.10 Furthermore, plasma glutamine was inversely correlated with indexes of obesity and dysglycemia in healthy Japanese adults,34 and a high plasma glutamine/glutamate ratio was associated with lower risk of incident diabetes mellitus in the FHS (Framingham Heart Study).35 In experimental models, administration of glutamine in mice led to both improved glucose tolerance and lower blood pressure.35 Therefore, in the context of these prior studies, our findings may indicate that visceral obesity reflects a relative “glutamine deficiency,” representing dysmetabolic, dysfunctional adiposity with adverse cardiometabolic consequences.

Lactate is a by‐product of anaerobic metabolism in cells when the energy‐producing capacity of aerobic metabolism is exceeded or when oxygen is not available to participate in cellular respiration. There is substantial evidence, particularly from animal studies, that hypoxia develops in adipose tissue as the tissue mass expands, and the reduction in the oxygen content underlies an inflammatory response.36 In hypoxic adipose tissue, secretion of multiple inflammation‐related adipokines is upregulated, and there is a switch from oxidative metabolism to anaerobic glycolysis, with corresponding increases in lactate production.37 The positive association between elevated lactate and VAT seen in our study may reflect the systemic effects of adipose tissue hypoxia, which are more common in VAT compared with other depots.38 Alternatively, higher lactate, seen in our study, may reflect abnormal mitochondrial function.39, 40 Metabolic flux studies using biological tracers have shown that glucose feeds the tricarboxylic acid cycle (an integral component of oxidative phosphorylation in the electron transport chain in mitochondria) via circulating lactate and that circulatory turnover flux of lactate is the highest of all metabolites, exceeding that of glucose in mice.41 Downregulation of several genes in the electron transport chain was found in viscerally obese women with diabetes mellitus and was, in part, mediated by expression of tumor necrosis factor‐α, an important inflammatory cytokine implicated in the pathogenesis of type 2 diabetes mellitus.42 A separate study also found that mitochondrial biogenesis and markers essential to aerobic metabolism were downregulated in acquired obesity in monozygotic twins.43 Furthermore, studies of inborn errors of metabolism related to mitochondrial dysfunction have identified multiple metabolites downstream of primary mitochondrial lesions, including lactate and several amino acids.44, 45 Therefore, alterations in whole body mitochondrial oxidative phosphorylation capacity in multiple tissues, reflected by metabolomics disturbances, may contribute to a shared pathogenesis of VAT accumulation and cardiometabolic disease.

Branched‐chain amino acids have been consistently linked to obesity and metabolic disease in recent years. Branched‐chain amino acids are activators of the mammalian target of rapamycin signaling pathway, and high concentrations of these amino acids induce mammalian target of rapamycin hyperactivity, leading to impaired pancreatic β cell insulin secretion and insulin resistance.46 Newgard and colleagues showed, in a rat model, that a dietary pattern of high‐fat consumption with branched‐chain amino acid supplementation led to obesity‐associated insulin resistance via long‐term activation of mammalian target of rapamycin that was reversed by the mammalian target of rapamycin inhibitor, rapamycin.47 They also used principal components analysis to show that branched‐chain amino acid concentrations can be used to differentiate metabolic signatures between obese and lean humans. Wang and colleagues further translated these findings to humans in the FHS by demonstrating that a branched‐chain amino acid signature was associated with elevated BMI48 and increased risk of type 2 diabetes mellitus.8 However, they found considerable overlap in metabolic profiles between BMI, insulin resistance, and dyslipidemia. Indeed, many studies have found similar “metabolic profiles” associated with a broad range of diseases, from diabetes mellitus to CVD, suggesting that alterations in the metabolic processes reflected by these biomarkers may be more indicative of generalized metabolic derangements rather than markers of a specific disease.49 Our findings may elucidate the reason for this metabolic overlap because excess visceral adiposity is a fundamental link between obesity and several adverse cardiometabolic traits.

It is well known that VAT is associated with an atherogenic, dyslipidemic lipid/lipoprotein profile, including high triglycerides, low HDL‐C,50, 51 smaller LDL and HDL particle size, larger VLDL size, and increased LDL and VLDL particle number.3 Indeed, in our study, HDL‐C, larger HDL‐related particles, and plasma apolipoprotein‐A1 (a major protein component of HDL particles in plasma) were inversely associated with VAT, whereas triglycerides and VLDL‐related particles were consistently positively associated with VAT (in both derivation and replication cohorts). Abnormalities in triglycerides and VLDL are more closely linked with entities classically related to VAT, such as the metabolic syndrome, insulin resistance, and the hypertriglyceridemic waist,52, 53 whereas alterations in HDL metabolism likely relate to atherogenesis through different mechanisms.54, 55 Therefore, our results may reflect multiple mechanistic pathways through which VAT and lipid/lipoproteins interact to influence cardiovascular and metabolic risk.

The ability to identify individuals before the onset of obesity‐related complications is particularly important for cardiometabolic diseases because therapies exist that can slow or prevent end‐organ damage over time. Although anthropometric indexes of obesity (eg, BMI and waist circumference) are easy to implement clinically, their correlation with direct imaging‐based assessments of visceral adiposity is modest; furthermore, these indexes incorporate both the abdominal subcutaneous and visceral depots that, as discussed, are anatomically and functionally distinct. Newer imaging‐based methods offer more sensitivity and specificity for measuring VAT but have significant drawbacks, limiting their use in clinical practice. Blood‐based metabolic profiling of VAT using a limited set of important metabolites may address this implementation gap between recognizing the role of visceral adiposity in cardiometabolic disease and actually assessing it clinically. Additional studies examining the relationship between metabolite signatures and future diabetes mellitus and/or cardiovascular events are an exciting next step in this field. Given that these new analyses would be more clinically oriented and require rigorous analytical approaches to evaluate the utility of metabolites in risk prediction for cardiometabolic events, they are beyond the scope of the current study.

In conclusion, from a panel of >30 000 metabolomics features, acetylglycoproteins, branched‐chain amino acids, lactate, glutamine, and markers of atherogenic dyslipidemia emerged as strong markers of visceral adiposity. A single, fasting measurement of these metabolites may provide additional information over standard risk markers of visceral fat (BMI, fasting glucose, waist circumference, and serum triglycerides). Further investigation is warranted to determine whether NMR‐based metabolic profiling can improve screening and detection of visceral adiposity beyond simple anthropometric measures and the hypertriglyceridemic waist to help identify appropriate candidates for interventions and reduce the cardiometabolic complications of visceral obesity.

Sources of Funding

Dr Neeland is supported by grant K23 DK106520 from the National Institutes of Health (NIH) and by the Dedman Family Scholarship in Clinical Care from the University of Texas Southwestern. MESA (Multi‐Ethnic Study of Atherosclerosis) was supported by contracts HHSN268201500003I, N01‐HC‐95159, N01‐HC‐95160, N01‐HC‐95161, N01‐HC‐95162, N01‐HC‐95163, N01‐HC‐95164, N01‐HC‐95165, N01‐HC‐95166, N01‐HC‐95167, N01‐HC‐95168, and N01‐HC‐95169 from the National Heart, Lung, and Blood Institute (NHLBI); and by grants UL1‐TR‐000040, UL1‐TR‐001079, and UL1‐TR‐001420 from the National Center for Advancing Translational Sciences. Dr Allison was supported by funding for MESA Abdominal Body Composition Ancillary study from the NHLBI (R01‐HL088451). Dr Karaman acknowledges support from the European Union PhenoMeNal Project (Horizon 2020, 654241). The Development of Combinatorial Biomarkers for Subclinical Atherosclerosis project was supported by a grant from the European Union Seventh Framework Programme (305422). The NEO (Netherlands Epidemiology in Obesity) study is supported by the participating departments, the division, and the Board of Directors of the Leiden University Medical Center, and by the Leiden University, Research Profile Area “Vascular and Regenerative Medicine.” We acknowledge support from the Netherlands Cardiovascular Research Initiative: an initiative with support of the Dutch Heart Foundation (CVON2014‐02 ENERGISE). Dr Mook‐Kanamori is supported by the Dutch Science Organization (ZonMW‐VENI Grant 916.14.023). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Disclosures

Dr Neeland has received honoraria, consulting and speaker's bureau fees, and travel support from Boehringer‐Ingelheim/Lilly Alliance (significant), has received a research grant from Novo Nordisk (significant), and is a member of the scientific advisory board of AMRA Medical (modest). The remaining authors have no disclosures to report.

Supporting information

Data S1. Supplemental Methods

Table S1. Metabolomics Biomarkers Measured in the NEO Study

Table S2. Biochemical Pathways Identified From Metabolomics Analysis Associated With Visceral Adiposity

Table S3. Spearman Correlation Coefficients for Body Mass Index, Visceral Adipose Tissue, and Selected Metabolites in MESA and NEO Studies

Figure S1. Metabolites associated with visceral adiposity in sex stratified analyses.

Figure S2. Metabolites associated with visceral adiposity in race/ethnicity stratified analyses.

Figure S3. Mendelian randomization study of genetic traits linked to blood lipid levels with visceral adiposity.

(J Am Heart Assoc. 2019;8:e010810 DOI: 10.1161/JAHA.118.010810.)